-
Survival analysis is a statistical method for studying individual survival probability and its response time. It is generally expressed as follows:
$$ P({t}_{i})=P(T={t}_{i})(i=1, 2, {...} ,k)$$ (1) where
$P\left({{t_i}} \right)$ is a survival function of the probability that migrant i resides longer than exit time t; T is the length of residence after migrant i has moved to a city. When T is a discrete random datum, the survival function of the exit time t for migrant i is expressed as:$$S{(}{t_i}{)} = P(T > t) = \sum\nolimits_{{t_i}} {P({t_i})} $$ (2) where S is the survival and P() is a probability function. Equation (2) is the final probability function, and the curve corresponding to the Equation (2) is a survival curve with a decreasing slope. We use two sets of microscale survey data—CMDS and CLDS data—to obtain the survival curve of China’s urban floating population. The CMDS data reflect the time points at which migrants moved to different cities but do not indicate when these migrants moved out of the cities; thus, we also use the CLDS data, which indicate when the migrants moved out of the cities. In the survival function, the exit time t, obtained from the CLDS data, is the time point of the last migration of a migrant.
To define the long-term residence of the urban floating population, we identify the inflection point of the survival curve at which the slope changes from large to small. In general, most of China’s urban floating population move out of a city within 1–3 yr; therefore, the survival probability of the urban floating population should decrease sharply within 1–3 yr. However, as time goes on, the deceleration rate of the survival probability slows down; the inflection point of the survival curve is thus crucial to maintaining a stable survival probability (Lin and Zhu, 2016). This inflection point is employed as the critical year to define the duration of the long-term residence among the urban floating population in China.
-
The classic approach to analyzing factors underlying migration destination choices is based on the random utility maximization framework (Cadwallader, 1989; Davies et al., 2001). In Cadwallader’s (1989) framework, micro and macro factors for migration destination choices are studied simultaneously. Micro factors include individual income, individual characteristics, and familial migration factors, where individual income and individual factors are the basis for a migrants decision to move (Tong and Wang, 2015), and the number of family members moving with a migrant also affects migration destination (Stark and Levhari, 1982; Chen and Rosenthal, 2008). Macro factors include the destination’s economy and amenities because the decision to move is generally based on a rational consideration of the migration destination (Liu and Shen, 2014; Liu and Xu, 2017).
Thus, the utility of a migrant in a city Uij consists of individual labor-force utility Ui, family labor-force utility Uh, and urban utility Uj. the utility model could be interpreted as follows:
$${U_{ij}}{\rm{ = }}F({U_i},{U_h},{U_j})$$ (3) where Uij is the utility function U of migrant i of household h in city j; Ui includes individual income and individual characteristics (e.g., age, sex, and education); Uh contains the number of family members moving with the migrant, household income level and expenditure level; and Uj comprises urban socioeconomic and physical environments (Cadwallader, 1989).
The aforementioned utility maximization framework could not be used to interpret long-term resident behavior, but it provided a valuable reference for the selection of basic factors in this study. The long-term residential behavior of a migrant involves a spatial equilibrium between cities, and the utility of individuals should be equal across cities in the long-term residence condition (Glaeser and Gottlieb, 2009). Therefore, the spatial equilibrium framework should also be considered in the analysis of factors underlying long-term residential behavior. On the basis of the fundamental work by Rosenand Roback, Glaeser (2011) proposed an urban simultaneous spatial equilibrium model. These spatial equilibrium models reflect the respective relationships of three explanatory variables (productivity, consumer amenities, and land areas) with three response variables (population density, wages, and housing prices), which are expressed as Equations (4)–(6).
$${\rm{ln(}}PD{)} = {\alpha _1} + {\beta _1}{\rm{ln(}}LS{)} + {\gamma _1}{\rm{ln(}}UA{)} + {\delta _1}{\rm{ln(}}LA{)}$$ (4) $${\rm{ln(}}LW{)} = {\alpha _2} + {\beta _2}{\rm{ln(}}LS{)} + {\gamma _2}{\rm{ln(}}UA{)} + {\delta _2}{\rm{ln(}}LA{)}$$ (5) $${\rm{ln(}}HP{)} = {\alpha _3} + {\beta _3}{\rm{ln(}}LS{)} + {\gamma _3}{\rm{ln(}}UA{)} + {\delta _3}{\rm{ln(}}LA{)}$$ (6) where
$PD,LS,UA,LA,LW,HP$ refer to population density, labor skills, urban amenities, land areas, labor wage, housing price, respectively. Generally, the response variable LW in Equation (5) equals to the individual income factor in Equation (3). Therefore, the individual income factor in Uij can be determined by urban factors including population density, housing prices, and skills. Population density (Glaeser and Shapiro, 2003; Glaeser and Gottlieb, 2009; Chen et al., 2018), consumption environment (Glaeser et al., 2001) and housing prices (Glaeser et al., 2006) are significant determinants for urban population and economic growth. Therefore, the final spatial utility function of migrant i in city j can be expressed as shown in Equation (7).$$ {U}_{ij}=F(PD, LW, HP, LS, UA, IC, HE)$$ (7) where
$IC,HE$ refer to individual character and household elements respectively. Equation (7) is the basic model of this study; it is a combination of the random utility maximization and spatial equilibrium frameworks. This model can not only fully consider the core factors of the utility maximization framework but also be used to interpret the long-term residential behavior of the urban floating population. Furthermore, the model derived by the simultaneous calculation of Equations (4)–(6) greatly reduced problems of endogeneity produced by the interaction between macro and micro explanatory factors.We further assume that the spatial utility Uij in Equation (7) is composed of a series of determinable factors and a random disturbance term, which is expressed as Equation (8).
$$ {U_{ij}} = \theta {x_{ij}} + {\varepsilon _{ij}} $$ (8) where
${x_{ij}},\theta,{\varepsilon _{ij}}$ is a series of determinable factor combinations, the coefficient vector, the random disturbance term, respectively. Because i and j are discrete variables, the probability that migrant i chooses long-termresidence is the probability that the spatial utility function${U_{ij}}$ is greater than 0. If the random disturbance term${\varepsilon _{ij}}$ obeys the normal distribution, then the probit regression model of long-term residence of migrant i in city j can be expressed as Equation (9).$$P{(}{\rm{choice}} = 1{)} = P\left[ {{U_{ij}} > 0} \right] = P\left[ {{\varepsilon _{ij}} < \theta {x_{ij}}} \right] = F{(}\theta {x_{ij}}{)}$$ (9) where
$F()$ is the cumulative distribution function of the standard normal distribution, and$P{(}{\rm{choice}} = 1{)}$ refers to the probability that migrant i is long-term resident in city j.Apart from the description of the basic model, the selection of macroscale indicators should be discussed. First, per capita GDP, the teacher-student ratio in colleges, and the average number of theaters are critical indicators for measuring urban economies, skills, and amenities. Second, population density and housing price indexes are essential in urban agglomeration economies (Glaeser et al., 2001; Duranton and Puga, 2014), which were not widely recognized until 2000.
Population density is typically useful for measuring the positive effect of urban agglomeration and the degree of urban crowding (Ciccone, 1996). In terms of the vulnerable urban floating population, the population density or population size index increases before it decreases (inverted U shape; Henderson, 1986). The vulnerable urban floating population in China is more sensitive to urban crowding than to the positive effect of agglomeration (Chan and Zhang, 1999). What’s more, during urban population agglomeration, crucial positive effects are mostly caused by the ‘learning effect’ (Duranton and Puga, 2014), which is only strong in megacities (Xi et al., 2019; Yu et al., 2019). The housing price index generally reflects the cost of living in cities, and this index can significantly reduce the residential behavior of the urban floating population (Liu et al., 2017). The housing price index also reflects megacity consumption environments, which can increase urban population growth (Glaeser et al., 2001; Glaeser and Gottlieb, 2006). The adverse effect of housing prices on the agglomeration of the vulnerable floating population is weaker in more prosperous cities because migrant residents prefer to rent rather than buy houses or apartments (Qin et al., 2018).
For these reasons, this study proposes the following two preliminary assumptions. First, population density negatively affects the long-term residence of the floating population moving to small cities while positively affects that of the floating population moving to megacities. Second, housing price negatively affects the long-term residence of the floating population moving to small cities with lower housing prices. If housing prices reach the level where the migrants can not afford the house, the migrants will be more likely to rent houses than buying them. Therefore, we assume that the effects of population density and housing price on the long-term residence of urban floating population are negative first and then positive.
-
The sources of microscale data in this paper are the China Migrants Dynamic Survey (CMDS) from 2012 to 2014, and China Labor-force Dynamics Survey (CLDS) in 2014 and 2016. CMDS is a large-scale national survey conducted by the National Health Commission once a year and focuses on the floating population aged 16–59 who has granted permission. The methods of this survey are stratified, multi-stage and sampling with probability proportional to size, and its annual sample size are near 200 000. CLDS conducted by Center for Social Survey of Sun Yat-sen University is a biennial tracking survey on the labor force aged 15–64, and has completed a collection of 3 yr (2012, 2014 and 2016). This dataset is an open-source data to all the teachers and students of Sun Yat-sen University, and we acquired this dataset in 2017 as PHD candidates. Macroscale data excluding Hongkong, Macao and Taiwan of China are obtained from the China Urban Statistical Yearbook and China Regional Economic Statistical Yearbook and Defense Meteorological Satellite Program–Operational Linescan System (DMSP/OLS) Nighttime Lights data (Yao et al., 2018; Chen et al., 2018).
The advantages of the microscale data employed lie in its two sources and long-time span. The CMDS is a large sample data set with more than 200 000 respondents from 262 prefecture-level cities, enabling the spatial pattern of the long-term residence of China’s urban floating population to be studied comprehensively. However, the CMDS data do not reveal when migrants moved out of their city. To clarify the long-term residence situation of the urban floating population, we further use the supplementary data of the CLDS. The advantage of CLDS data is that it includes the historical migration trajectory of migrants, making it possible to apply survival analysis. However, CLDS contains the defects of small sample size and limited coverage of cities, which could fully reflect the spatial pattern of long-term residence of the urban floating population in China. Overall, the application of both data sets ensures the accuracy and rationality of the research results.
Moreover, we match the microscale data of migrants with the macroeconomic data of cities to emphasize the effects of urban factors on the residence of the floating population. Because of considerable difficulty in data collection, we employ macroscale data for only 2010 and 2013; the macroeconomic data for 2010 are used to represent the situation in 2012, whereas the data of 2013 represent the situation of 2013, 2014, and 2016. As mentioned previously, the factors employed herein to represent prefecture-level cities, including population density index, per capita GDP, urban housing price index, number of opera houses, and teacher-student ratio in universities. Population density is the permanent urban population size divided by the area of built-upon land, and the size of the permanent urban population is the sum of the registered population and temporary population. Registered population data are obtained from the China City Statistical Yearbook (2011; 2014) (NBSC, 2011; 2014) while temporary population data from the China Urban Construction Statistical Yearbook (2010; 2013) (MOHURD, 2010; 2013). Urban construction land data are obtained from the DMSP/OLS Nighttime Lights by using the spatiotemporal normalized threshold method. This dataset is obtained by dividing the national night light data into more than 2800 NTL images at county scale based on the TM 30 meters data in 2010, which can greatly improve the accuracy of the urban construction land area data (Chen et al., 2018; Yao et al., 2018). Urban housing price data are extracted from China Statistical Yearbook for Regional Economy (2011; 2014) (NBSC, 2011; 2014). Other data including the number of opera houses, teacher-student ratio in universities are extracted from the China City Statistical Yearbook (2011; 2014) (NBSC, 2011; 2014). In order to clarify the spatial pattern of long-term residence, we choose the criterion of the four economic regions processed by National Bureau of Statistics (NBSC, 2013). The four economic regions are eastern China, central China, western China and northeastern China. Eastern China consists of Beijing, Tianjin, Hebei, Shanghai, Jiangsu, Zhejiang, Fujian, Shandong, Guangdong and Hainan. Central China consists of Shanxi, Anhui, Jiangxi, Henan, Hubei, Hunan. Western China consists of Inner Mongolia, Guangxi, Chongqing, Sichuan, Guizhou, Yunnan, Tibet, Shannxi, Gansu, Qinghai, Ningxia and Xinjiang. Northeastern China consists of Liaoning, Jilin and Heilongjiang. The study doesn’t include Taiwan, Hong Kong and Macao.
We process the original microscale survey data as follows. For CMDS data, we consider 262 prefecture-level cities where statistics are available for the period 2012–2014; thus, contrastive study of different years could be performed. Cities not included in CMDS 2012 are Chaoyang, Mudanjiang, Laiwu, Maoming, Guang’an, Ya’an, Yichun, Zaozhuang, Taian, Ezhou, Suizhou, Shanwei, and Jieyang. Cities excluded from CMDS 2013 are Quzhou, Heze, Hebi, Xuchang, Shangqiu, Zhoukou, Jiyuan, Meizhou, and Yangjiang. Cities not studied in CMDS 2014 are Fuxin, Qitaihe, Quzhou, Hebi, Xuchang, Shangqiu, Zhoukou, Jiyuan, Meizhou, Yangjiang, and Yunfu. For CLDS data, we select the survey years 2014 and 2016 and employ historical data of the last migration of migrants from 2008 to 2016. After this data cleaning, the CMDS dataset in 2012, 2013, and 2014 have contained data on 142 021, 179 512, and 179 963 individuals, respectively, whereas the CLDS dataset in 2014 and 2016 contain data on 2217 and 1,369 people respectively. The summary of the statistics of key variables is presented in Table 1.
CMDS 2012 2013 2014 N Mean SD N Mean SD N Mean SD Length of residence 136870 4.321 (4.520) 171964 4.500 (4.554) 172084 4.409 (4.547) Gender 136870 0.525 (0.499) 171964 0.534 (0.499) 172084 0.579 (0.494) Age 136870 33.402 (9.208) 171964 33.631 (9.178) 172084 33.846 (9.245) Edu 136870 9.803 (2.764) 171964 9.863 (2.716) 172084 10.110 (2.864) Ln (population density) 133072 8.689 (0.592) 165684 8.573 (0.580) 166687 8.580 (0.575) Ln (housing price) 136870 8.736 (0.635) 171964 8.842 (0.513) 172084 8.842 (0.511) Ln (urban income) 136870 10.660 (0.596) 165684 10.813 (0.679) 166687 10.832 (0.662) Number of theatre 136471 31.603 (44.14) 171844 32.613 (57.15) 171924 32.952 (57.080) Teacher student ratio 135594 0.063 (0.014) 171964 0.060 (0.017) 172084 0.060 (0.017) Table 1. Statistics of the key variables
CLDS 2014 2016 N Mean SD N Mean SD Length of residence 2117 6.107 (4.160) 1369 6.604 (4.700) Gender 2117 0.443 (0.497) 1369 0.457 (0.498) Age 2102 34.624 (11.240) 1363 35.158 (11.248) Edu 2116 10.635 (3.801) 1368 10.613 (3.883) Ln (population density) 2117 8.795 (0.517) 1369 8.830 (0.491) Ln (housing price) 2117 8.748 (0.652) 1369 8.787 (0.634) Ln (urban income) 2117 10.726 (0.656) 1369 10.794 (0.623) Number of theatre 2112 26.212 (31.801) 1362 28.853 (36.992) Teacher student ratio 2117 0.142 (0.074) 1369 0.145 (0.078) Notes: The education variable is expressed by the actual years of education of the labor force. For the mean of age, the China Migrants Dynamic Survey (CMDS) data corresponds to the mean age of the floating population aged 16–59, and the China Labor-force Dynamics Survey (CLDS) data corresponds to the mean age of the labor force aged 15–64. The unit of population density is persons/km2; housing price is yuan (RMB)/km2; and urban income, expressed in per capita GDP, yuan(RMB)/person. -
Using the two robust sets of microscale survey data, we obtain the survival probability curve of the floating population (Figs. 1 and 2). The inflection points of the curves, at which the slopes of the survival curve change from large to small, are identified and used as the criteria for defining long-term residence of China’s urban floating population. We discover that ‘5 yr’ corresponds to the inflection point at which China’s urban floating population changes status from short-term to long-term residence; that is, those living in a city for six or more years are identified as long-term residents.
Figure 1. Determining the inflection point of the survival curves obtained using China Migrants Dynamic Survey (CMDS) data on the long-term residence of the floating population. Figs. 1a, 1b, and 1c are the survival curves obtained using 2012, 2013, and 2014 data, respectively. Fig. 1d shows the proportion of the floating population that stayed in their city for the indicated periods
Figure 2. Survival curve of China’s floating population based on China Labor-force Dynamics Survey (CLDS) data. Figs. 2a and 2b are the survival curves of urban residence of the floating population in 2014 and 2016, respectively. This graph is obtained using the Kaplan–Meier method of survival analysis
Fig. 1 presents the survival curve of the long-term residence of China’s urban floating population obtained through the CMDS dataset from 2012 to 2014. The curves suggest that 5 yr is the inflection point. Specifically, Fig. 1a shows the survival curves based on the 2012 CMDS data. This figure meets the criterion for the basic form of a survival curve, being a smooth convex curve enclosing the origin with the slope decreasing. Crucially, the inflection point is found to be located at both 4 and 5 yr, and the survival probability of a migrant staying in a city for less than or equal to 4 yr is greater than 40%, whereas the survival probability of a migrant staying in a city for more than or equal to 6 yr is less than 25%. Similarly, Figs. 1b and 1c present the survival curves based on CMDS data for 2013 and 2014, respectively. The inflection point in these figures is solely at 5 yr. To verify the accuracy of the results of Figs. 1a–1c, we plotted Fig. 1d, which illustrates the proportion of the total floating population that stayed in their city for 0–1 yr, 1 yr, 2 yr, …, up to 10 yr. Fig. 1d confirms that 5 yr corresponds to the inflection point at which the proportion of the total floating population becomes similar from year to year. In the first five years of residence of the floating population, the average decline rate of the survival proportion is 22.2%, whereas during the 6–10 yr after the inflow, the average decline rate is only 15.9%.
Fig. 2 displays the survival curve obtained using the CLDS data for 2014 and 2016. The research results again indicate that 5 yr is the inflection point. Specifically, Fig. 2a is the survival curve obtained using CLDS data for 2014, and we found that the slope of the survival curve changes from large to small at 5 yr. The average decline rate of the survival probability from 1 yr to 5 yr is 5.4%, whereas the average decline rate from 6 to 10 yr is only 2.9%. Similar to Figs. 2a, 2b shows the survival curve obtained using CLDS data for 2016 and indicates that the inflection point is between 5 and 6 yr.
Compared with those in Fig. 1, the survival probabilities plotted in Fig. 2 are considerably higher. This is probably because the CLDS data used to plot Fig. 2 reflect the outflow condition of the urban floating population; that is, the curve estimates the actual outflow of migrant i after t yr of residence in city j. Thus, compared with the curve in Fig. 1, which neglects the outflow of the urban floating population, the survival probability in Fig. 2 is more accurate. Otherwise, the CMDS data employed to obtain Fig. 1 include data for 0–1 yr. By contrast, the CLDS data employed to obtain Fig. 2 reflect the survival probability from the first year after beginning residence; thus, the exit probability of residence of 0–1 yr is coded as being that for 0 yr.
-
We wish to determine the spatial pattern of the long-term residence of China’s urban floating population. Using the index of the proportion of the floating population that has resided in a city for 6 yr or more, we analyze this proportion in each prefecture-level city and then obtain the spatial pattern of long-term residence of China’s urban floating population. The results show that the regions in which migrants are more likely to be long-term residents are the megacities in the three urban agglomerations in eastern China (Yangtze River Delta, Pearl River Delta, and Beijing-Tianjin-Hebei Urban Agglomerations) as well as small and medium-sized cities in western China and northeastern China; by contrast, the short-term residence areas are more commonly the cities in central China and near the three urban agglomerations in eastern China.
Fig. 3 displays the spatial distribution of long-term residence that was obtained using CMDS data for 2012–2014. Three crucial findings are obtained through the analysis of the special patterns of 2012, 2013 and 2014. First, the three urban agglomerations in eastern China are common long-term residence regions of the urban floating population in China. Second, cities in central China and near the three large metropolitan areas are the regions in which migrants have the shortest residence. Third, cities in western China and northeastern China, especially small and medium-sized cities, are also important long-term residence regions of China’s urban floating population.
Figure 3. Spatial pattern of long-term residence of China’s urban floating population, obtained from China Migrants Dynamic Survey (CMDS) data without Hong Kong, Macao and Taiwan of China. Figs. 3a, 3b, and 3c show the spatial pattern in 2012, 2013, and 2014, respectively. The spatial pattern displayed in Fig. 3d is based on the proportion of the floating population residing for 6 yr or more during 2012–1014
Fig. 3a depicts the spatial distribution obtained using CMDS data for 2012. The proportion of the floating population residing in the three urban agglomerations urban agglomerations and cities in western China and northeastern China who are long-term residents of these cities is higher than 28% (i.e., less than 72% of these residents are short-term residents), whereas for cities in central China and near the large three urban agglomerations in eastern China, this proportion is less than 24%. Specifically, the average proportion of the total floating population who are long-term residents in one of the three urban agglomerations is 27.9%, and those for the migrants who are long-term residents in the first-tier cities—Beijing, Shanghai, Guangzhou, and Shenzhen—are higher at 39%, 45%, 32%, and 30% respectively. Regarding the western cities, the average proportion is 28.8%, and the traditionally labor exporting cities and immigrant cities have high rates of long-term residence. For example, Dazhou, traditionally a labor exporting city, has 36% of the floating population who are long-term residents, whereas the corresponding percentage for Panzhihua, a traditional immigrant city, is 37%. In addition, the proportion of cities in northeastern China is even higher at 37.3%, with those of Shuangyashan and Heihe being 52% and 60%, respectively. However, the long-term-resident proportion of the urban floating population in cities near the large three urban agglomerationsurban agglomerations in eastern China and in the cities in central China is only 23.6% and 22.8%, respectively. The short-term residence characteristics of the urban floating population are particularly prominent for some provincial capitals and large-scale cities, such as Jinan, Hefei, and Fuyang, for which the proportions are only 14%, 13%, and 6%, respectively.
Figs. 3b and 3c depict the spatial distribution pattern obtained using the 2013 and 2014 data. They also show that the three urban agglomerations and the cities in western China and northeastern China are notable long-term residence areas, with proportions of long-term residence of 29%, 36%, and 31% respectively. By contrast, cities in central China and other cities in eastern China are common short-term residence areas, with proportions of 24% and 25%, respectively. We also depict the proportion of long-term residents among the floating population for each city from 2012 to 2014 (Fig. 3d) and find that this spatial distribution pattern is consistent with those in Figs. 3a–3c.
Using the CMDS data for 2012–2014 and CLDS data for 2014 and 2016, we further calculate the average time of residence of the floating population by region and city size (Table 2). The results prove the accuracy of the findings obtained from Fig. 3. Regarding the CMDS data, the average residence time is longer than 4.6 yr in the three urban agglomerations and cities in northeastern China, 4.5 yr in cities in western China, and only 4 yr in cities in central China and other cities in eastern China. Simultaneously, the average residence time is longer than 4.8 yr in megacities, 4.5 yr in small and medium-sized cities, and less than 4 yr in big cities. We obtain the same results using the CLDS data. Notably, however, the probable reason for the difference of average resident time shown by CLDS and CMDS is that the CLDS data contain the outflow information of the urban floating population, which is usually larger than the data that do not contain the outflow information.
Classification Different types of cities CMDS CLDS 2012 2013 2014 average 2014 2016 By region Three Urban Agglomerations in Eastern China 4.86 4.86 4.71 4.81 5.62 5.94 Cities in Eastern China without Three Urban Agglomerations 3.83 4.18 4.00 4.00 5.34 5.42 Cities in Central China 3.69 4.15 4.22 4.02 5.42 5.27 Cities in Northeastern China 4.67 5.03 4.94 4.88 5.96 5.97 Cities in Western China 4.21 4.60 4.60 4.47 6.14 5.74 By city size Megacities (> 10 million residents) 5.08 4.98 4.83 4.96 5.52 5.95 Big cities (5–10 million residents) 3.70 4.00 3.88 3.86 5.38 5.74 Medium cities (3–5 million residents) 4.34 4.59 4.68 4.64 5.54 5.83 Small cities (< 3 million residents) 4.38 4.87 5.09 4.78 6.07 6.03 Notes: The average data for the China Migrants Dynamic Survey (CMDS) are based on the average residence time of the total floating population in each region from 2012 to 2014. Data sources: China Migrants Dynamic Survey (CMDS) 2012–2014; China Labor-force Dynamics Survey (CLDS) 2014 and 2016 Table 2. Spatial differences in average residence time of urban floating population in China
-
Table 3 reports the regression results obtained using Equation (9) and the CMDS data. Urban density and urban housing price are discovered to both exert a ‘U-shaped’ influence on long-term residence of the urban floating population, whereas urban income, number of theaters per capita, and teacher-student ratio have significantly positive effects in all models. Columns 1 and 2 in Table 3 present the regression results for the 2012 CMDS data. As expected, the coefficients for population density and housing price are significantly negative and their quadratic coefficients are positive, indicating that the influences of population density and housing price on long-term residence are U-shaped. Additionally, urban income, number of theaters, and teacher-student ratio have significant positive impacts. Specifically, in column 1, the coefficient of urban population density and its square term are –0.386 and 0.01, respectively.
Elements Explanatory variables 2012 2013 2014 (1) (2) (3) (4) (5) (6) Urban elements Population density –0.386*** –0.211** –0.501*** –0.378*** –0.499*** –0.353*** (–2.973) (–2.179) (–8.066) (–6.438) (–8.027) (–5.953) Population density square 0.010*** 0.006** 0.026*** 0.019*** 0.025*** 0.017*** (2.676) (2.073) (6.942) (5.439) (6.657) (4.772) Housing price –0.850*** –0.787*** –1.131*** –0.854*** –1.381*** –1.178*** (–13.433) (–12.947) (–13.665) (–10.924) (–17.072) (–15.229) Housing price square 0.051*** 0.047*** 0.064*** 0.049*** 0.076*** 0.065*** (14.060) (13.610) (13.743) (11.199) (16.788) (15.060) Urban income 0.017*** 0.012*** 0.075*** 0.059*** 0.097*** 0.080*** (4.625) (3.272) (18.478) (15.459) (25.181) (21.728) Number of theatre per capita 0.128** 0.016*** 0.128* 0.119 0.253** 0.217 (2.572) (3.341) (1.932) (1.566) (2.247) (1.349) Teacher student ratio 1.787*** 1.598*** 0.867*** 0.848*** 1.157*** 1.174*** (17.952) (16.724) (11.062) (11.461) (14.756) (15.641) Micro individual elements Gender –0.003 0.002 0.013*** (–1.288) (0.941) (6.050) Age 0.010*** 0.011*** 0.010*** (67.111) (87.423) (76.867) Marriage –0.054*** –0.038*** –0.064*** (–9.427) (–7.564) (–12.627) Edu –0.004*** –0.004*** –0.004*** (–9.775) (–9.308) (–10.920) Micro familial elements Household income 0.020*** 0.007** 0.006** (7.261) (2.556) (2.097) Household Expenditure 0.072*** 0.075*** 0.073*** (25.702) (28.285) (28.424) Only spouse migration 0.043*** 0.061*** 0.046*** (7.844) (12.950) (9.312) Only children migration 0.011 0.037*** 0.023** (1.185) (4.274) (2.572) Nuclear family migration 0.115*** 0.130*** 0.123*** (22.106) (29.444) (26.502) N 129407 128255 165564 164026 166527 166310 Adjusted R2 0.084 0.106 0.078 0.102 0.087 0.094 Notes: t statistics are in parentheses; ***, **, * denote statistical significance at 1%, 5%, and 10% level, respectively. Data source: China Migrants Dynamic Survey (CMDS) 2012–2014 Table 3. Explanation of long-term residence based on probit regression
When controlling the microscale individual and familial elements (column 2), the effects of the two variables are largely the same in terms of significance level and sign, with the coefficient of urban population density and its square being −0.211 and 0.006, respectively, and the coefficient of urban housing price and its square term being −0.787 and 0.047, respectively. Columns 3 and 4 and columns 5 and 6 present the results for 2013 and 2014, respectively. The sign and significance of the core variables are relatively consistent, supporting the conclusion that the effects of urban density and housing price are U-shaped and that of urban agglomeration factors is significantly positive. Regarding the control variables, compared with individual elements, familial elements, especially household income and number of family members moving together, make a more significant contribution to the long-term residence of the floating population.
For the CLDS data, Table 4 reports the results obtained using Cox and Weibull regression, which support the findings presented in Table 3. Columns 1 and 2 show the results of the Cox model, with column 1 showing the results for model including only macroscale urban elements. For this model, urban density and urban housing price are found to exert U-shaped effects on long-term residence, with coefficients and quadratic coefficients of −0.783, −0.161 and 0.050, 0.022 respectively. The coefficients for urban income and number of theaters per capita are again positive and highly significant, whereas the coefficient for teacher-student ratio is significantly negative, which indicates that the positive influence of the college and university element is not robust. The results obtained for the models including microscale individual elements as control variables are presented in column 2, and the sign and significance of most of the urban variables are consistent except for the coefficient of urban density, which is nonsignificant. Columns 3 and 4 show the results obtained using Weibull regression, which is generally consistent with those obtained using the Cox method. Thus, the findings in Table 4 verify that the effects of urban population density and urban housing price on long-term residence are significantly U-shaped, whereas the impact coefficients of urban income and urban consumption environment are significantly positive.
Explanatory variables Cox Weibull (1) (2) (3) (4) Population density –0.783** –0.524 –0.844*** –0.582 (–2.553) (–1.501) (–2.715) (–1.625) Population density square 0.050*** 0.035** 0.052*** 0.037** (3.308) (2.042) (3.422) (2.125) Housing price –0.161*** –0.298** –0.148** –0.276* (–3.630) (–2.122) (–2.003) (–1.865) Housing price square 0.022*** 0.022** 0.021*** 0.019** (3.404) (2.099) (3.350) (2.086) Urban income 0.443*** 0.325* 0.463*** 0.366* (2.795) (1.706) (2.891) (1.880) Number of theatre per capita 0.539*** 0.670*** 0.562*** 0.720*** (10.104) (9.171) (10.653) (9.799) Teacher-student ratio –4.239*** –4.576*** –4.214*** –4.626*** (–6.927) (–5.868) (–6.873) (–5.933) Control variables No Yes No Yes N 3090 1892 3090 1892 Notes: t statistics are in parentheses; ***, **, * denote statistical significance at 1%, 5%, and 10% level, respectively. Data source: China Labor-force Dynamics Survey (CLDS) 2014 and 2016 Table 4. Explanation of long-term residence based on survival analysis
Spatial Pattern of Long-term Residence in the Urban Floating Population of China and its Influencing Factors
doi: 10.1007/s11769-021-1193-9
- Received Date: 2020-02-15
- Accepted Date: 2020-06-14
- Publish Date: 2021-03-01
-
Key words:
- long-term residence /
- urban floating population /
- spatial pattern /
- spatial utility equilibrium model /
- China
Abstract: Exploring long-term residence among the urban floating population is crucial to understanding urban growth in China, particularly since the 2008 financial crisis. By using China Migrants Dynamic Survey data for 2012–2014, China Labor-force Dynamics Survey data for 2014–2016, and macroscale urban matched data, we analyzed the spatial pattern of long-term residential behavior in China’s urban floating population in 2012–2016 and developed an urban spatial utility equilibrium model containing ‘macro’ urban factors and ‘micro’ individual and household factors to explain the pattern. The results first revealed that long-term residence is defined as ≥ 6 yr for the urban floating population in China. Second, members of this population are more likely to be long-term residents of the megacities in the three urban agglomerations in eastern China as well as of small and medium-sized cities in western and northeastern China, whereas short-term residence is more likely in cities in central China and near the three urban agglomerations. Third, urban population density and housing prices, both have a significant U-shaped effect, are main factors affecting the spatial pattern of long-term residence.
Citation: | CHEN Le, XI Meijun, JIN Wanfu, HU Ya, 2021. Spatial Pattern of Long-term Residence in the Urban Floating Population of China and its Influencing Factors. Chinese Geographical Science, 31(2): 342−358 doi: 10.1007/s11769-021-1193-9 |